Skip to content

Dataset · MBZUAI

Multi-speaker vocalized Arabic speech corpus

MBZUAI/ArVoice

MBZUAI Modern Standard Arabic audio dataset with multiple speakers and full diacritics in transcripts.

What's inside

The corpus contains 10K to 100K audio-text pairs in Modern Standard Arabic (MSA) recorded by various speakers. The key feature is the complete placement of diacritic marks (vocalization) in the text transcripts. The dataset was prepared by MBZUAI researchers.

What you can do with it

  • Training multi-speaker text-to-speech (TTS) models in Arabic
  • Training and testing automatic speech recognition (ASR) systems considering diacritics
  • Voice conversion tasks, speech diacritics restoration, and deepfake detection
Tasks
Text → speech, Text → audio, Automatic speech recognition
Languages
Arabic
Size
10K–100K rows
Downloads this month
1.5K
Likes
34

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("MBZUAI/ArVoice")